AI tooling: what's drawing attention

The saved-reading backlog from recent weeks clusters around a single thread: ChatGPT, OpenAI's chat-style interface that answers requests in back-and-forth exchanges. It sits alongside other AI-flavored tools — GitHub's CoPilot for code and Dall-E for generated images and art — and raises the usual questions: the end of human development, a new way to make art, or just cocktail-party material?

Several pieces probe what the model actually does when put to work:

  • A Conversation With ChatGPT (Matthias Ott) — Ott discusses typography with the model, pushing into theoretical design-process territory. Asked whether designers should learn to code, it answers that this is a decision each designer must make for themselves, based on their own goals and circumstances; some may benefit from coding, others from concentrating on design principles and concepts.
  • They were supposed to replace the creative jobs last (Dave Rupert) — Rupert doubts large gains arrive without realized costs, electricity among them, and wonders whether the price paid is our perception of reality itself: the next thing you read or watch is increasingly likely to come out of a content extruder.
  • I just used ChatGPT to help with a complicated equation (Reddit) — a user had it write a complex equation in Notion; a couple of hiccups, but it worked.
  • ChatGPT Creates a Working WordPress Plugin – On the First Try (WP Tavern) — Sarah Gooding covers Johnathon Williams producing a fully functional WordPress plugin from a simple chat command, a result both terrifying and mind-blowing.
  • ChatGPT Is a Smart Computer's Impression of a Know-It-All (Pixel Envy) — Nick Heer points to a Atlantic article opening with three paragraphs written by ChatGPT, natural enough even if slightly fishy on first read.
  • Stack Overflow: temporary ban — the site temporarily banned ChatGPT-generated text in content, amid users posting model-produced code as answers.
  • Midjourney vs. human illustrators: has AI already won? (Evil Martians) — the experiment is a clear reminder that AI doesn't simply work: at best it performs like a junior designer, and after two and a half hours of back-and-forth the author gave up and upscaled the most promising result. The post closes with realistic situations where AI could help a team — not replace a whole person's job.
  • Quick Thoughts on AI (Collaborative Fund) — includes a chart of how fast ChatGPT reached one million users: ten months for Facebook, five days here.

Testing the image side

A prompt sent to Dall-E asked for a photo of a developer at a desk, head exploding while chatting with an AI algorithm on a desktop computer.

Not bad, not bad.